62 → 117 / 144
Golden-corpus accuracy
← All case studies
AI & agentic workflows Internal work sample
Controlled cross-ERP entity matching
Raised golden-corpus matching accuracy from 62 to 117 of 144 while admitting zero hard-negative automatic accepts, creating a controlled path towards a reliable cross-ERP master.
2,426
Pairs re-scored
185
Entity scope
62 Baseline scorer
weighted fuzzy scoring only
117 Calibrated bands with LLM referee
zero hard-negative auto-accepts
Scores from the documented golden-corpus evaluation (144 labelled pairs), 2026.
Evidence inputs
- Entity records across heterogeneous ERP systems with inconsistent names, spellings and legal forms.
- Python · rapidfuzz · scikit-learn (TF-IDF)
Transformation
- Normalised legal-form noise before pre-filtering candidates with TF-IDF and weighted fuzzy matching.
- Sent only the ambiguous residual to a model referee.
- Banded every pair into automatic accept, human review or no match.
Controls & assurance
- Set the automatic-accept threshold for zero hard-negative false accepts, accepting more review to avoid silent wrong merges.
- Kept scores, thresholds and final merge decisions deterministic and auditable.
Output
- Golden-corpus accuracy rose from 62 to 117 of 144 with no hard-negative auto-accepts; a production sweep re-scored 2,426 pairs into the three bands for review.
Business value
- Made a reliable cross-ERP master achievable and identified the master-data clean-up required before a group CRM go-live.
- Problem
Different ERP systems held the same entities under inconsistent names and no common key, blocking clean consolidation.
- Approach
Normalised legal-form noise before pre-filtering candidates with TF-IDF and weighted fuzzy matching.
- Outcome
Golden-corpus accuracy rose from 62 to 117 of 144 with no hard-negative auto-accepts; a production sweep re-scored 2,426 pairs into the three bands for review.
Business value
Made a reliable cross-ERP master achievable and identified the master-data clean-up required before a group CRM go-live.
Transformation route
- 01
Normalised legal-form noise before pre-filtering candidates with TF-IDF and weighted fuzzy matching.
- 02
Sent only the ambiguous residual to a model referee.
- 03
Banded every pair into automatic accept, human review or no match.
Decision log
- Set the automatic-accept threshold for zero hard-negative false accepts, accepting more review to avoid silent wrong merges.
- Kept scores, thresholds and final merge decisions deterministic and auditable.
This case study proves
- AI coding and implementation assistants AI & Automation Applied
- Master-data reconciliation / entity resolution Data & Reporting Strong
What I learned
- Calibrated bands and a review lane protect control integrity better than a higher raw match rate with silent errors.
Future improvements
- Expand the labelled corpus and track precision and recall by band as a regression control.